Otwarty dostęp

Classification of emotions based on electrodermal activity and transfer learning - a pilot study

Fredrik A. Jacobsen
Fredrik A. Jacobsen
, 
Ellen W. Hafli
Ellen W. Hafli
, 
Christian Tronstad
Christian Tronstad
 oraz 
Ørjan G. Martinsen
Ørjan G. Martinsen
   | 30 gru 2021
Journal of Electrical Bioimpedance's Cover Image
O artykule
Poprzedni artykuł
Następny artykuł
Zacytuj
Data publikacji: 30 gru 2021
Zakres stron: 178 - 183
Otrzymano: 07 gru 2021
DOI: https://doi.org/10.2478/joeb-2021-0021
Słowa kluczowe
© 2021 Fredrik A. Jacobsen, Ellen W. Hafli, Christian Tronstad, Ørjan G. Martinsen, published by Sciendo
This work is licensed under the Creative Commons Attribution 4.0 International License.

Figure 1

Example measurement of low frequency skin conductance. Person being highly stressed (red curve), moderately stressed (green curve), and totally relaxed (blue curve).
Example measurement of low frequency skin conductance. Person being highly stressed (red curve), moderately stressed (green curve), and totally relaxed (blue curve).

Figure 2

Conductance measurement. Example of data reported as disgust.
Conductance measurement. Example of data reported as disgust.

Figure 3

CWT treated conductance measurement. The same example (Figure 2) of data reported as disgust with CWT applied.
CWT treated conductance measurement. The same example (Figure 2) of data reported as disgust with CWT applied.

Figure 4

Architecture of the machine learning process.
Architecture of the machine learning process.

Figure 5

Confusion matrix for the first results produced by the model using the full dataset.
Confusion matrix for the first results produced by the model using the full dataset.

Figure 6

Confusion matrix for the first results produced by the model using the full data set with SMOTE data added.
Confusion matrix for the first results produced by the model using the full data set with SMOTE data added.

Figure 7

Confusion matrix for the results produced by the model using the three category data set with SMOTE data added.
Confusion matrix for the results produced by the model using the three category data set with SMOTE data added.

Results using the full CWT applied EDA dataset with synthetic data. The parameters are set to: test size = 20 and trees = 100.

test size = 20 trees = 100
precision recall f1-score support
amusement 1.00 0.50 0.67 4
anger 0.60 0.75 0.67 4
disgust 0.33 0.25 0.29 4
fear 1.00 1.00 1.00 6
neutral 0.80 0.44 0.57 9
sadness 0.27 0.60 0.37 5
accuracy 0.59 32
macro avg 0.67 0.59 0.59 32
weighted avg 0.70 0.59 0.61 32

Results using the 3 class CWT EDA data with synthetic data. Test size = 20 and trees = 300.

test size = 20 trees = 600
precision recall f1-score support
amusement 1.00 0.80 0.89 5
disgust 0.80 0.67 0.73 6
sadness 0.71 1.00 0.83 5
accuracy 0.81 16
macro avg 0.84 0.82 0.82 16
weighted avg 0.84 0.81 0.81 16

Results using the CWT treated EDA data with test size = 20 and trees = 100.

test size = 20 trees = 100
precision recall f1-score support
amusement 0.75 0.75 0.75 4
anger 0.00 0.00 0.00 2
disgust 1.00 0.25 0.40 8
fear 0.00 0.00 0.00 1
neutral 0.00 0.00 0.00 2
sadness 0.15 0.67 0.25 3
accuracy 0.35 20
macro avg 0.32 0.28 0.23 20
weighted avg 0.57 0.35 0.35 20

Category of emotion and how many times the emotion was reported.

Emotion Number of samples
Amusement 20
Anger 9
Disgust 24
Fear 5
Neutral 15
Sadness 26
Tenderness 1
eISSN:
1891-5469
Język:
Angielski
Częstotliwość wydawania:
Volume Open
Dziedziny czasopisma:
Engineering, Bioengineering and Biomedical Engineering, Biomedical Electronics, Life Sciences, Biophysics, Medicine, Biomedical Engineering, Physics, Spectroscopy and Metrology
Kanał RSS czasopisma